]>
Commit | Line | Data |
---|---|---|
3dcf60bc | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
1da177e4 LT |
3 | * Block device elevator/IO-scheduler. |
4 | * | |
5 | * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
6 | * | |
0fe23479 | 7 | * 30042000 Jens Axboe <axboe@kernel.dk> : |
1da177e4 LT |
8 | * |
9 | * Split the elevator a bit so that it is possible to choose a different | |
10 | * one or even write a new "plug in". There are three pieces: | |
11 | * - elevator_fn, inserts a new request in the queue list | |
12 | * - elevator_merge_fn, decides whether a new buffer can be merged with | |
13 | * an existing request | |
14 | * - elevator_dequeue_fn, called when a request is taken off the active list | |
15 | * | |
16 | * 20082000 Dave Jones <davej@suse.de> : | |
17 | * Removed tests for max-bomb-segments, which was breaking elvtune | |
18 | * when run without -bN | |
19 | * | |
20 | * Jens: | |
21 | * - Rework again to work with bio instead of buffer_heads | |
22 | * - loose bi_dev comparisons, partition handling is right now | |
23 | * - completely modularize elevator setup and teardown | |
24 | * | |
25 | */ | |
26 | #include <linux/kernel.h> | |
27 | #include <linux/fs.h> | |
28 | #include <linux/blkdev.h> | |
29 | #include <linux/elevator.h> | |
30 | #include <linux/bio.h> | |
1da177e4 LT |
31 | #include <linux/module.h> |
32 | #include <linux/slab.h> | |
33 | #include <linux/init.h> | |
34 | #include <linux/compiler.h> | |
2056a782 | 35 | #include <linux/blktrace_api.h> |
9817064b | 36 | #include <linux/hash.h> |
0835da67 | 37 | #include <linux/uaccess.h> |
c8158819 | 38 | #include <linux/pm_runtime.h> |
eea8f41c | 39 | #include <linux/blk-cgroup.h> |
1da177e4 | 40 | |
55782138 LZ |
41 | #include <trace/events/block.h> |
42 | ||
242f9dcb | 43 | #include "blk.h" |
bd166ef1 | 44 | #include "blk-mq-sched.h" |
bca6b067 | 45 | #include "blk-pm.h" |
8330cdb0 | 46 | #include "blk-wbt.h" |
242f9dcb | 47 | |
1da177e4 LT |
48 | static DEFINE_SPINLOCK(elv_list_lock); |
49 | static LIST_HEAD(elv_list); | |
50 | ||
9817064b JA |
51 | /* |
52 | * Merge hash stuff. | |
53 | */ | |
83096ebf | 54 | #define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq)) |
9817064b | 55 | |
da775265 JA |
56 | /* |
57 | * Query io scheduler to see if the current process issuing bio may be | |
58 | * merged with rq. | |
59 | */ | |
72ef799b | 60 | static int elv_iosched_allow_bio_merge(struct request *rq, struct bio *bio) |
da775265 | 61 | { |
165125e1 | 62 | struct request_queue *q = rq->q; |
b374d18a | 63 | struct elevator_queue *e = q->elevator; |
da775265 | 64 | |
f9cd4bfe JA |
65 | if (e->type->ops.allow_merge) |
66 | return e->type->ops.allow_merge(q, rq, bio); | |
da775265 JA |
67 | |
68 | return 1; | |
69 | } | |
70 | ||
1da177e4 LT |
71 | /* |
72 | * can we safely merge with this request? | |
73 | */ | |
72ef799b | 74 | bool elv_bio_merge_ok(struct request *rq, struct bio *bio) |
1da177e4 | 75 | { |
050c8ea8 | 76 | if (!blk_rq_merge_ok(rq, bio)) |
72ef799b | 77 | return false; |
7ba1ba12 | 78 | |
72ef799b TE |
79 | if (!elv_iosched_allow_bio_merge(rq, bio)) |
80 | return false; | |
1da177e4 | 81 | |
72ef799b | 82 | return true; |
1da177e4 | 83 | } |
72ef799b | 84 | EXPORT_SYMBOL(elv_bio_merge_ok); |
1da177e4 | 85 | |
68c43f13 DLM |
86 | static inline bool elv_support_features(unsigned int elv_features, |
87 | unsigned int required_features) | |
8ac0d9a8 | 88 | { |
68c43f13 DLM |
89 | return (required_features & elv_features) == required_features; |
90 | } | |
91 | ||
92 | /** | |
93 | * elevator_match - Test an elevator name and features | |
94 | * @e: Scheduler to test | |
95 | * @name: Elevator name to test | |
96 | * @required_features: Features that the elevator must provide | |
97 | * | |
5b8f65e1 RD |
98 | * Return true if the elevator @e name matches @name and if @e provides all |
99 | * the features specified by @required_features. | |
68c43f13 DLM |
100 | */ |
101 | static bool elevator_match(const struct elevator_type *e, const char *name, | |
102 | unsigned int required_features) | |
103 | { | |
104 | if (!elv_support_features(e->elevator_features, required_features)) | |
105 | return false; | |
8ac0d9a8 JA |
106 | if (!strcmp(e->elevator_name, name)) |
107 | return true; | |
108 | if (e->elevator_alias && !strcmp(e->elevator_alias, name)) | |
109 | return true; | |
110 | ||
111 | return false; | |
112 | } | |
113 | ||
68c43f13 DLM |
114 | /** |
115 | * elevator_find - Find an elevator | |
116 | * @name: Name of the elevator to find | |
117 | * @required_features: Features that the elevator must provide | |
118 | * | |
119 | * Return the first registered scheduler with name @name and supporting the | |
120 | * features @required_features and NULL otherwise. | |
2527d997 | 121 | */ |
68c43f13 DLM |
122 | static struct elevator_type *elevator_find(const char *name, |
123 | unsigned int required_features) | |
1da177e4 | 124 | { |
a22b169d | 125 | struct elevator_type *e; |
1da177e4 | 126 | |
70cee26e | 127 | list_for_each_entry(e, &elv_list, list) { |
68c43f13 | 128 | if (elevator_match(e, name, required_features)) |
a22b169d | 129 | return e; |
1da177e4 | 130 | } |
1da177e4 | 131 | |
a22b169d | 132 | return NULL; |
1da177e4 LT |
133 | } |
134 | ||
135 | static void elevator_put(struct elevator_type *e) | |
136 | { | |
137 | module_put(e->elevator_owner); | |
138 | } | |
139 | ||
2527d997 JA |
140 | static struct elevator_type *elevator_get(struct request_queue *q, |
141 | const char *name, bool try_loading) | |
1da177e4 | 142 | { |
2824bc93 | 143 | struct elevator_type *e; |
1da177e4 | 144 | |
2a12dcd7 | 145 | spin_lock(&elv_list_lock); |
2824bc93 | 146 | |
68c43f13 | 147 | e = elevator_find(name, q->required_elevator_features); |
21c3c5d2 | 148 | if (!e && try_loading) { |
e1640949 | 149 | spin_unlock(&elv_list_lock); |
490b94be | 150 | request_module("%s-iosched", name); |
e1640949 | 151 | spin_lock(&elv_list_lock); |
68c43f13 | 152 | e = elevator_find(name, q->required_elevator_features); |
e1640949 JA |
153 | } |
154 | ||
2824bc93 TH |
155 | if (e && !try_module_get(e->elevator_owner)) |
156 | e = NULL; | |
157 | ||
2a12dcd7 | 158 | spin_unlock(&elv_list_lock); |
1da177e4 LT |
159 | return e; |
160 | } | |
161 | ||
3d1ab40f AV |
162 | static struct kobj_type elv_ktype; |
163 | ||
d50235b7 | 164 | struct elevator_queue *elevator_alloc(struct request_queue *q, |
165125e1 | 165 | struct elevator_type *e) |
3d1ab40f | 166 | { |
b374d18a | 167 | struct elevator_queue *eq; |
9817064b | 168 | |
c1b511eb | 169 | eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, q->node); |
9817064b | 170 | if (unlikely(!eq)) |
8406a4d5 | 171 | return NULL; |
9817064b | 172 | |
22f746e2 | 173 | eq->type = e; |
f9cb074b | 174 | kobject_init(&eq->kobj, &elv_ktype); |
9817064b | 175 | mutex_init(&eq->sysfs_lock); |
242d98f0 | 176 | hash_init(eq->hash); |
9817064b | 177 | |
3d1ab40f AV |
178 | return eq; |
179 | } | |
d50235b7 | 180 | EXPORT_SYMBOL(elevator_alloc); |
3d1ab40f AV |
181 | |
182 | static void elevator_release(struct kobject *kobj) | |
183 | { | |
b374d18a | 184 | struct elevator_queue *e; |
9817064b | 185 | |
b374d18a | 186 | e = container_of(kobj, struct elevator_queue, kobj); |
22f746e2 | 187 | elevator_put(e->type); |
3d1ab40f AV |
188 | kfree(e); |
189 | } | |
190 | ||
c3e22192 | 191 | void __elevator_exit(struct request_queue *q, struct elevator_queue *e) |
1da177e4 | 192 | { |
3d1ab40f | 193 | mutex_lock(&e->sysfs_lock); |
dd1c372d | 194 | blk_mq_exit_sched(q, e); |
3d1ab40f | 195 | mutex_unlock(&e->sysfs_lock); |
1da177e4 | 196 | |
3d1ab40f | 197 | kobject_put(&e->kobj); |
1da177e4 | 198 | } |
2e662b65 | 199 | |
9817064b JA |
200 | static inline void __elv_rqhash_del(struct request *rq) |
201 | { | |
242d98f0 | 202 | hash_del(&rq->hash); |
e8064021 | 203 | rq->rq_flags &= ~RQF_HASHED; |
9817064b JA |
204 | } |
205 | ||
70b3ea05 | 206 | void elv_rqhash_del(struct request_queue *q, struct request *rq) |
9817064b JA |
207 | { |
208 | if (ELV_ON_HASH(rq)) | |
209 | __elv_rqhash_del(rq); | |
210 | } | |
bd166ef1 | 211 | EXPORT_SYMBOL_GPL(elv_rqhash_del); |
9817064b | 212 | |
70b3ea05 | 213 | void elv_rqhash_add(struct request_queue *q, struct request *rq) |
9817064b | 214 | { |
b374d18a | 215 | struct elevator_queue *e = q->elevator; |
9817064b JA |
216 | |
217 | BUG_ON(ELV_ON_HASH(rq)); | |
242d98f0 | 218 | hash_add(e->hash, &rq->hash, rq_hash_key(rq)); |
e8064021 | 219 | rq->rq_flags |= RQF_HASHED; |
9817064b | 220 | } |
bd166ef1 | 221 | EXPORT_SYMBOL_GPL(elv_rqhash_add); |
9817064b | 222 | |
70b3ea05 | 223 | void elv_rqhash_reposition(struct request_queue *q, struct request *rq) |
9817064b JA |
224 | { |
225 | __elv_rqhash_del(rq); | |
226 | elv_rqhash_add(q, rq); | |
227 | } | |
228 | ||
70b3ea05 | 229 | struct request *elv_rqhash_find(struct request_queue *q, sector_t offset) |
9817064b | 230 | { |
b374d18a | 231 | struct elevator_queue *e = q->elevator; |
b67bfe0d | 232 | struct hlist_node *next; |
9817064b JA |
233 | struct request *rq; |
234 | ||
ee89f812 | 235 | hash_for_each_possible_safe(e->hash, rq, next, hash, offset) { |
9817064b JA |
236 | BUG_ON(!ELV_ON_HASH(rq)); |
237 | ||
238 | if (unlikely(!rq_mergeable(rq))) { | |
239 | __elv_rqhash_del(rq); | |
240 | continue; | |
241 | } | |
242 | ||
243 | if (rq_hash_key(rq) == offset) | |
244 | return rq; | |
245 | } | |
246 | ||
247 | return NULL; | |
248 | } | |
249 | ||
2e662b65 JA |
250 | /* |
251 | * RB-tree support functions for inserting/lookup/removal of requests | |
252 | * in a sorted RB tree. | |
253 | */ | |
796d5116 | 254 | void elv_rb_add(struct rb_root *root, struct request *rq) |
2e662b65 JA |
255 | { |
256 | struct rb_node **p = &root->rb_node; | |
257 | struct rb_node *parent = NULL; | |
258 | struct request *__rq; | |
259 | ||
260 | while (*p) { | |
261 | parent = *p; | |
262 | __rq = rb_entry(parent, struct request, rb_node); | |
263 | ||
83096ebf | 264 | if (blk_rq_pos(rq) < blk_rq_pos(__rq)) |
2e662b65 | 265 | p = &(*p)->rb_left; |
796d5116 | 266 | else if (blk_rq_pos(rq) >= blk_rq_pos(__rq)) |
2e662b65 | 267 | p = &(*p)->rb_right; |
2e662b65 JA |
268 | } |
269 | ||
270 | rb_link_node(&rq->rb_node, parent, p); | |
271 | rb_insert_color(&rq->rb_node, root); | |
2e662b65 | 272 | } |
2e662b65 JA |
273 | EXPORT_SYMBOL(elv_rb_add); |
274 | ||
275 | void elv_rb_del(struct rb_root *root, struct request *rq) | |
276 | { | |
277 | BUG_ON(RB_EMPTY_NODE(&rq->rb_node)); | |
278 | rb_erase(&rq->rb_node, root); | |
279 | RB_CLEAR_NODE(&rq->rb_node); | |
280 | } | |
2e662b65 JA |
281 | EXPORT_SYMBOL(elv_rb_del); |
282 | ||
283 | struct request *elv_rb_find(struct rb_root *root, sector_t sector) | |
284 | { | |
285 | struct rb_node *n = root->rb_node; | |
286 | struct request *rq; | |
287 | ||
288 | while (n) { | |
289 | rq = rb_entry(n, struct request, rb_node); | |
290 | ||
83096ebf | 291 | if (sector < blk_rq_pos(rq)) |
2e662b65 | 292 | n = n->rb_left; |
83096ebf | 293 | else if (sector > blk_rq_pos(rq)) |
2e662b65 JA |
294 | n = n->rb_right; |
295 | else | |
296 | return rq; | |
297 | } | |
298 | ||
299 | return NULL; | |
300 | } | |
2e662b65 JA |
301 | EXPORT_SYMBOL(elv_rb_find); |
302 | ||
34fe7c05 CH |
303 | enum elv_merge elv_merge(struct request_queue *q, struct request **req, |
304 | struct bio *bio) | |
1da177e4 | 305 | { |
b374d18a | 306 | struct elevator_queue *e = q->elevator; |
9817064b | 307 | struct request *__rq; |
06b86245 | 308 | |
488991e2 AB |
309 | /* |
310 | * Levels of merges: | |
311 | * nomerges: No merges at all attempted | |
312 | * noxmerges: Only simple one-hit cache try | |
313 | * merges: All merge tries attempted | |
314 | */ | |
7460d389 | 315 | if (blk_queue_nomerges(q) || !bio_mergeable(bio)) |
488991e2 AB |
316 | return ELEVATOR_NO_MERGE; |
317 | ||
9817064b JA |
318 | /* |
319 | * First try one-hit cache. | |
320 | */ | |
72ef799b | 321 | if (q->last_merge && elv_bio_merge_ok(q->last_merge, bio)) { |
34fe7c05 CH |
322 | enum elv_merge ret = blk_try_merge(q->last_merge, bio); |
323 | ||
06b86245 TH |
324 | if (ret != ELEVATOR_NO_MERGE) { |
325 | *req = q->last_merge; | |
326 | return ret; | |
327 | } | |
328 | } | |
1da177e4 | 329 | |
488991e2 | 330 | if (blk_queue_noxmerges(q)) |
ac9fafa1 AB |
331 | return ELEVATOR_NO_MERGE; |
332 | ||
9817064b JA |
333 | /* |
334 | * See if our hash lookup can find a potential backmerge. | |
335 | */ | |
4f024f37 | 336 | __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector); |
72ef799b | 337 | if (__rq && elv_bio_merge_ok(__rq, bio)) { |
9817064b | 338 | *req = __rq; |
866663b7 ML |
339 | |
340 | if (blk_discard_mergable(__rq)) | |
341 | return ELEVATOR_DISCARD_MERGE; | |
9817064b JA |
342 | return ELEVATOR_BACK_MERGE; |
343 | } | |
344 | ||
f9cd4bfe JA |
345 | if (e->type->ops.request_merge) |
346 | return e->type->ops.request_merge(q, req, bio); | |
1da177e4 LT |
347 | |
348 | return ELEVATOR_NO_MERGE; | |
349 | } | |
350 | ||
5e84ea3a JA |
351 | /* |
352 | * Attempt to do an insertion back merge. Only check for the case where | |
353 | * we can append 'rq' to an existing request, so we can throw 'rq' away | |
354 | * afterwards. | |
355 | * | |
fd2ef39c JK |
356 | * Returns true if we merged, false otherwise. 'free' will contain all |
357 | * requests that need to be freed. | |
5e84ea3a | 358 | */ |
fd2ef39c JK |
359 | bool elv_attempt_insert_merge(struct request_queue *q, struct request *rq, |
360 | struct list_head *free) | |
5e84ea3a JA |
361 | { |
362 | struct request *__rq; | |
bee0393c | 363 | bool ret; |
5e84ea3a JA |
364 | |
365 | if (blk_queue_nomerges(q)) | |
366 | return false; | |
367 | ||
368 | /* | |
369 | * First try one-hit cache. | |
370 | */ | |
fd2ef39c JK |
371 | if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq)) { |
372 | list_add(&rq->queuelist, free); | |
5e84ea3a | 373 | return true; |
fd2ef39c | 374 | } |
5e84ea3a JA |
375 | |
376 | if (blk_queue_noxmerges(q)) | |
377 | return false; | |
378 | ||
bee0393c | 379 | ret = false; |
5e84ea3a JA |
380 | /* |
381 | * See if our hash lookup can find a potential backmerge. | |
382 | */ | |
bee0393c SL |
383 | while (1) { |
384 | __rq = elv_rqhash_find(q, blk_rq_pos(rq)); | |
385 | if (!__rq || !blk_attempt_req_merge(q, __rq, rq)) | |
386 | break; | |
387 | ||
fd2ef39c | 388 | list_add(&rq->queuelist, free); |
bee0393c SL |
389 | /* The merged request could be merged with others, try again */ |
390 | ret = true; | |
391 | rq = __rq; | |
392 | } | |
27419322 | 393 | |
bee0393c | 394 | return ret; |
5e84ea3a JA |
395 | } |
396 | ||
34fe7c05 CH |
397 | void elv_merged_request(struct request_queue *q, struct request *rq, |
398 | enum elv_merge type) | |
1da177e4 | 399 | { |
b374d18a | 400 | struct elevator_queue *e = q->elevator; |
1da177e4 | 401 | |
f9cd4bfe JA |
402 | if (e->type->ops.request_merged) |
403 | e->type->ops.request_merged(q, rq, type); | |
06b86245 | 404 | |
2e662b65 JA |
405 | if (type == ELEVATOR_BACK_MERGE) |
406 | elv_rqhash_reposition(q, rq); | |
9817064b | 407 | |
06b86245 | 408 | q->last_merge = rq; |
1da177e4 LT |
409 | } |
410 | ||
165125e1 | 411 | void elv_merge_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
412 | struct request *next) |
413 | { | |
b374d18a | 414 | struct elevator_queue *e = q->elevator; |
bd166ef1 | 415 | |
f9cd4bfe JA |
416 | if (e->type->ops.requests_merged) |
417 | e->type->ops.requests_merged(q, rq, next); | |
06b86245 | 418 | |
9817064b | 419 | elv_rqhash_reposition(q, rq); |
06b86245 | 420 | q->last_merge = rq; |
1da177e4 LT |
421 | } |
422 | ||
165125e1 | 423 | struct request *elv_latter_request(struct request_queue *q, struct request *rq) |
1da177e4 | 424 | { |
b374d18a | 425 | struct elevator_queue *e = q->elevator; |
1da177e4 | 426 | |
f9cd4bfe JA |
427 | if (e->type->ops.next_request) |
428 | return e->type->ops.next_request(q, rq); | |
bd166ef1 | 429 | |
1da177e4 LT |
430 | return NULL; |
431 | } | |
432 | ||
165125e1 | 433 | struct request *elv_former_request(struct request_queue *q, struct request *rq) |
1da177e4 | 434 | { |
b374d18a | 435 | struct elevator_queue *e = q->elevator; |
1da177e4 | 436 | |
f9cd4bfe JA |
437 | if (e->type->ops.former_request) |
438 | return e->type->ops.former_request(q, rq); | |
bd166ef1 | 439 | |
a1ce35fa | 440 | return NULL; |
1da177e4 LT |
441 | } |
442 | ||
3d1ab40f AV |
443 | #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr) |
444 | ||
445 | static ssize_t | |
446 | elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
1da177e4 | 447 | { |
3d1ab40f | 448 | struct elv_fs_entry *entry = to_elv(attr); |
b374d18a | 449 | struct elevator_queue *e; |
3d1ab40f AV |
450 | ssize_t error; |
451 | ||
452 | if (!entry->show) | |
453 | return -EIO; | |
454 | ||
b374d18a | 455 | e = container_of(kobj, struct elevator_queue, kobj); |
3d1ab40f | 456 | mutex_lock(&e->sysfs_lock); |
22f746e2 | 457 | error = e->type ? entry->show(e, page) : -ENOENT; |
3d1ab40f AV |
458 | mutex_unlock(&e->sysfs_lock); |
459 | return error; | |
460 | } | |
1da177e4 | 461 | |
3d1ab40f AV |
462 | static ssize_t |
463 | elv_attr_store(struct kobject *kobj, struct attribute *attr, | |
464 | const char *page, size_t length) | |
465 | { | |
3d1ab40f | 466 | struct elv_fs_entry *entry = to_elv(attr); |
b374d18a | 467 | struct elevator_queue *e; |
3d1ab40f | 468 | ssize_t error; |
1da177e4 | 469 | |
3d1ab40f AV |
470 | if (!entry->store) |
471 | return -EIO; | |
1da177e4 | 472 | |
b374d18a | 473 | e = container_of(kobj, struct elevator_queue, kobj); |
3d1ab40f | 474 | mutex_lock(&e->sysfs_lock); |
22f746e2 | 475 | error = e->type ? entry->store(e, page, length) : -ENOENT; |
3d1ab40f AV |
476 | mutex_unlock(&e->sysfs_lock); |
477 | return error; | |
478 | } | |
479 | ||
52cf25d0 | 480 | static const struct sysfs_ops elv_sysfs_ops = { |
3d1ab40f AV |
481 | .show = elv_attr_show, |
482 | .store = elv_attr_store, | |
483 | }; | |
484 | ||
485 | static struct kobj_type elv_ktype = { | |
486 | .sysfs_ops = &elv_sysfs_ops, | |
487 | .release = elevator_release, | |
488 | }; | |
489 | ||
cecf5d87 | 490 | int elv_register_queue(struct request_queue *q, bool uevent) |
3d1ab40f | 491 | { |
5a5bafdc | 492 | struct elevator_queue *e = q->elevator; |
3d1ab40f AV |
493 | int error; |
494 | ||
f0c6ae09 YY |
495 | lockdep_assert_held(&q->sysfs_lock); |
496 | ||
b2d6db58 | 497 | error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched"); |
3d1ab40f | 498 | if (!error) { |
22f746e2 | 499 | struct elv_fs_entry *attr = e->type->elevator_attrs; |
3d1ab40f | 500 | if (attr) { |
e572ec7e AV |
501 | while (attr->attr.name) { |
502 | if (sysfs_create_file(&e->kobj, &attr->attr)) | |
3d1ab40f | 503 | break; |
e572ec7e | 504 | attr++; |
3d1ab40f AV |
505 | } |
506 | } | |
cecf5d87 ML |
507 | if (uevent) |
508 | kobject_uevent(&e->kobj, KOBJ_ADD); | |
509 | ||
430c62fb | 510 | e->registered = 1; |
3d1ab40f AV |
511 | } |
512 | return error; | |
1da177e4 | 513 | } |
bc1c1169 | 514 | |
1da177e4 LT |
515 | void elv_unregister_queue(struct request_queue *q) |
516 | { | |
f0c6ae09 YY |
517 | lockdep_assert_held(&q->sysfs_lock); |
518 | ||
f8fc877d TH |
519 | if (q) { |
520 | struct elevator_queue *e = q->elevator; | |
521 | ||
522 | kobject_uevent(&e->kobj, KOBJ_REMOVE); | |
523 | kobject_del(&e->kobj); | |
cecf5d87 | 524 | |
f8fc877d | 525 | e->registered = 0; |
8330cdb0 JK |
526 | /* Re-enable throttling in case elevator disabled it */ |
527 | wbt_enable_default(q); | |
f8fc877d | 528 | } |
1da177e4 LT |
529 | } |
530 | ||
e567bf71 | 531 | int elv_register(struct elevator_type *e) |
1da177e4 | 532 | { |
e42cfb1d DLM |
533 | /* insert_requests and dispatch_request are mandatory */ |
534 | if (WARN_ON_ONCE(!e->ops.insert_requests || !e->ops.dispatch_request)) | |
535 | return -EINVAL; | |
536 | ||
3d3c2379 TH |
537 | /* create icq_cache if requested */ |
538 | if (e->icq_size) { | |
539 | if (WARN_ON(e->icq_size < sizeof(struct io_cq)) || | |
540 | WARN_ON(e->icq_align < __alignof__(struct io_cq))) | |
541 | return -EINVAL; | |
542 | ||
543 | snprintf(e->icq_cache_name, sizeof(e->icq_cache_name), | |
544 | "%s_io_cq", e->elevator_name); | |
545 | e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size, | |
546 | e->icq_align, 0, NULL); | |
547 | if (!e->icq_cache) | |
548 | return -ENOMEM; | |
549 | } | |
550 | ||
551 | /* register, don't allow duplicate names */ | |
2a12dcd7 | 552 | spin_lock(&elv_list_lock); |
68c43f13 | 553 | if (elevator_find(e->elevator_name, 0)) { |
3d3c2379 | 554 | spin_unlock(&elv_list_lock); |
62d2a194 | 555 | kmem_cache_destroy(e->icq_cache); |
3d3c2379 TH |
556 | return -EBUSY; |
557 | } | |
1da177e4 | 558 | list_add_tail(&e->list, &elv_list); |
2a12dcd7 | 559 | spin_unlock(&elv_list_lock); |
1da177e4 | 560 | |
d0b0a81a HT |
561 | printk(KERN_INFO "io scheduler %s registered\n", e->elevator_name); |
562 | ||
3d3c2379 | 563 | return 0; |
1da177e4 LT |
564 | } |
565 | EXPORT_SYMBOL_GPL(elv_register); | |
566 | ||
567 | void elv_unregister(struct elevator_type *e) | |
568 | { | |
3d3c2379 | 569 | /* unregister */ |
2a12dcd7 | 570 | spin_lock(&elv_list_lock); |
1da177e4 | 571 | list_del_init(&e->list); |
2a12dcd7 | 572 | spin_unlock(&elv_list_lock); |
3d3c2379 TH |
573 | |
574 | /* | |
575 | * Destroy icq_cache if it exists. icq's are RCU managed. Make | |
576 | * sure all RCU operations are complete before proceeding. | |
577 | */ | |
578 | if (e->icq_cache) { | |
579 | rcu_barrier(); | |
580 | kmem_cache_destroy(e->icq_cache); | |
581 | e->icq_cache = NULL; | |
582 | } | |
1da177e4 LT |
583 | } |
584 | EXPORT_SYMBOL_GPL(elv_unregister); | |
585 | ||
d48ece20 | 586 | int elevator_switch_mq(struct request_queue *q, |
54d5329d OS |
587 | struct elevator_type *new_e) |
588 | { | |
589 | int ret; | |
590 | ||
14a23498 BVA |
591 | lockdep_assert_held(&q->sysfs_lock); |
592 | ||
54d5329d | 593 | if (q->elevator) { |
b89f625e | 594 | if (q->elevator->registered) |
54d5329d | 595 | elv_unregister_queue(q); |
cecf5d87 | 596 | |
54d5329d OS |
597 | ioc_clear_queue(q); |
598 | elevator_exit(q, q->elevator); | |
599 | } | |
600 | ||
601 | ret = blk_mq_init_sched(q, new_e); | |
602 | if (ret) | |
603 | goto out; | |
604 | ||
605 | if (new_e) { | |
cecf5d87 | 606 | ret = elv_register_queue(q, true); |
54d5329d OS |
607 | if (ret) { |
608 | elevator_exit(q, q->elevator); | |
609 | goto out; | |
610 | } | |
611 | } | |
612 | ||
613 | if (new_e) | |
614 | blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name); | |
615 | else | |
616 | blk_add_trace_msg(q, "elv switch: none"); | |
617 | ||
618 | out: | |
54d5329d | 619 | return ret; |
54d5329d OS |
620 | } |
621 | ||
61db437d DLM |
622 | static inline bool elv_support_iosched(struct request_queue *q) |
623 | { | |
6251b754 | 624 | if (!queue_is_mq(q) || |
7a7c5e71 | 625 | (q->tag_set && (q->tag_set->flags & BLK_MQ_F_NO_SCHED))) |
61db437d DLM |
626 | return false; |
627 | return true; | |
628 | } | |
629 | ||
131d08e1 | 630 | /* |
a0958ba7 DLM |
631 | * For single queue devices, default to using mq-deadline. If we have multiple |
632 | * queues or mq-deadline is not available, default to "none". | |
633 | */ | |
634 | static struct elevator_type *elevator_get_default(struct request_queue *q) | |
635 | { | |
90b71980 BVA |
636 | if (q->tag_set && q->tag_set->flags & BLK_MQ_F_NO_SCHED_BY_DEFAULT) |
637 | return NULL; | |
638 | ||
580dca81 ML |
639 | if (q->nr_hw_queues != 1 && |
640 | !blk_mq_is_sbitmap_shared(q->tag_set->flags)) | |
a0958ba7 DLM |
641 | return NULL; |
642 | ||
643 | return elevator_get(q, "mq-deadline", false); | |
644 | } | |
645 | ||
646 | /* | |
647 | * Get the first elevator providing the features required by the request queue. | |
648 | * Default to "none" if no matching elevator is found. | |
649 | */ | |
650 | static struct elevator_type *elevator_get_by_features(struct request_queue *q) | |
651 | { | |
a2614255 | 652 | struct elevator_type *e, *found = NULL; |
a0958ba7 DLM |
653 | |
654 | spin_lock(&elv_list_lock); | |
655 | ||
656 | list_for_each_entry(e, &elv_list, list) { | |
657 | if (elv_support_features(e->elevator_features, | |
a2614255 JA |
658 | q->required_elevator_features)) { |
659 | found = e; | |
a0958ba7 | 660 | break; |
a2614255 | 661 | } |
a0958ba7 DLM |
662 | } |
663 | ||
a2614255 JA |
664 | if (found && !try_module_get(found->elevator_owner)) |
665 | found = NULL; | |
a0958ba7 DLM |
666 | |
667 | spin_unlock(&elv_list_lock); | |
a2614255 | 668 | return found; |
a0958ba7 DLM |
669 | } |
670 | ||
671 | /* | |
672 | * For a device queue that has no required features, use the default elevator | |
673 | * settings. Otherwise, use the first elevator available matching the required | |
674 | * features. If no suitable elevator is find or if the chosen elevator | |
675 | * initialization fails, fall back to the "none" elevator (no elevator). | |
131d08e1 | 676 | */ |
954b4a5c | 677 | void elevator_init_mq(struct request_queue *q) |
131d08e1 CH |
678 | { |
679 | struct elevator_type *e; | |
954b4a5c | 680 | int err; |
131d08e1 | 681 | |
61db437d | 682 | if (!elv_support_iosched(q)) |
954b4a5c | 683 | return; |
61db437d | 684 | |
75e6c00f | 685 | WARN_ON_ONCE(blk_queue_registered(q)); |
c48dac13 | 686 | |
131d08e1 | 687 | if (unlikely(q->elevator)) |
954b4a5c | 688 | return; |
131d08e1 | 689 | |
a0958ba7 DLM |
690 | if (!q->required_elevator_features) |
691 | e = elevator_get_default(q); | |
692 | else | |
693 | e = elevator_get_by_features(q); | |
131d08e1 | 694 | if (!e) |
954b4a5c | 695 | return; |
131d08e1 | 696 | |
737eb78e DLM |
697 | blk_mq_freeze_queue(q); |
698 | blk_mq_quiesce_queue(q); | |
699 | ||
131d08e1 | 700 | err = blk_mq_init_sched(q, e); |
737eb78e DLM |
701 | |
702 | blk_mq_unquiesce_queue(q); | |
703 | blk_mq_unfreeze_queue(q); | |
704 | ||
954b4a5c DLM |
705 | if (err) { |
706 | pr_warn("\"%s\" elevator initialization failed, " | |
707 | "falling back to \"none\"\n", e->elevator_name); | |
131d08e1 | 708 | elevator_put(e); |
954b4a5c | 709 | } |
131d08e1 | 710 | } |
131d08e1 | 711 | |
1da177e4 LT |
712 | /* |
713 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
714 | * we don't free the old io scheduler, before we have allocated what we | |
715 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 716 | * one, if the new one fails init for some reason. |
1da177e4 | 717 | */ |
165125e1 | 718 | static int elevator_switch(struct request_queue *q, struct elevator_type *new_e) |
1da177e4 | 719 | { |
e8989fae | 720 | int err; |
1da177e4 | 721 | |
14a23498 BVA |
722 | lockdep_assert_held(&q->sysfs_lock); |
723 | ||
a1ce35fa JA |
724 | blk_mq_freeze_queue(q); |
725 | blk_mq_quiesce_queue(q); | |
4722dc52 | 726 | |
a1ce35fa | 727 | err = elevator_switch_mq(q, new_e); |
1da177e4 | 728 | |
a1ce35fa JA |
729 | blk_mq_unquiesce_queue(q); |
730 | blk_mq_unfreeze_queue(q); | |
75ad23bc | 731 | |
5dd531a0 | 732 | return err; |
1da177e4 LT |
733 | } |
734 | ||
5dd531a0 JA |
735 | /* |
736 | * Switch this queue to the given IO scheduler. | |
737 | */ | |
7c8a3679 | 738 | static int __elevator_change(struct request_queue *q, const char *name) |
1da177e4 LT |
739 | { |
740 | char elevator_name[ELV_NAME_MAX]; | |
741 | struct elevator_type *e; | |
742 | ||
e9a823fb | 743 | /* Make sure queue is not in the middle of being removed */ |
58c898ba | 744 | if (!blk_queue_registered(q)) |
e9a823fb DJ |
745 | return -ENOENT; |
746 | ||
bd166ef1 JA |
747 | /* |
748 | * Special case for mq, turn off scheduling | |
749 | */ | |
fbd72127 AZ |
750 | if (!strncmp(name, "none", 4)) { |
751 | if (!q->elevator) | |
752 | return 0; | |
bd166ef1 | 753 | return elevator_switch(q, NULL); |
fbd72127 | 754 | } |
cd43e26f | 755 | |
ee2e992c | 756 | strlcpy(elevator_name, name, sizeof(elevator_name)); |
2527d997 | 757 | e = elevator_get(q, strstrip(elevator_name), true); |
340ff321 | 758 | if (!e) |
1da177e4 | 759 | return -EINVAL; |
1da177e4 | 760 | |
68c43f13 DLM |
761 | if (q->elevator && |
762 | elevator_match(q->elevator->type, elevator_name, 0)) { | |
2ca7d93b | 763 | elevator_put(e); |
5dd531a0 | 764 | return 0; |
2ca7d93b | 765 | } |
1da177e4 | 766 | |
5dd531a0 JA |
767 | return elevator_switch(q, e); |
768 | } | |
7c8a3679 | 769 | |
5dd531a0 JA |
770 | ssize_t elv_iosched_store(struct request_queue *q, const char *name, |
771 | size_t count) | |
772 | { | |
773 | int ret; | |
774 | ||
6251b754 | 775 | if (!elv_support_iosched(q)) |
5dd531a0 JA |
776 | return count; |
777 | ||
7c8a3679 | 778 | ret = __elevator_change(q, name); |
5dd531a0 JA |
779 | if (!ret) |
780 | return count; | |
781 | ||
5dd531a0 | 782 | return ret; |
1da177e4 LT |
783 | } |
784 | ||
165125e1 | 785 | ssize_t elv_iosched_show(struct request_queue *q, char *name) |
1da177e4 | 786 | { |
b374d18a | 787 | struct elevator_queue *e = q->elevator; |
bd166ef1 | 788 | struct elevator_type *elv = NULL; |
70cee26e | 789 | struct elevator_type *__e; |
1da177e4 LT |
790 | int len = 0; |
791 | ||
344e9ffc | 792 | if (!queue_is_mq(q)) |
cd43e26f MP |
793 | return sprintf(name, "none\n"); |
794 | ||
bd166ef1 JA |
795 | if (!q->elevator) |
796 | len += sprintf(name+len, "[none] "); | |
797 | else | |
798 | elv = e->type; | |
cd43e26f | 799 | |
2a12dcd7 | 800 | spin_lock(&elv_list_lock); |
70cee26e | 801 | list_for_each_entry(__e, &elv_list, list) { |
68c43f13 | 802 | if (elv && elevator_match(elv, __e->elevator_name, 0)) { |
1da177e4 | 803 | len += sprintf(name+len, "[%s] ", elv->elevator_name); |
bd166ef1 JA |
804 | continue; |
805 | } | |
68c43f13 DLM |
806 | if (elv_support_iosched(q) && |
807 | elevator_match(__e, __e->elevator_name, | |
808 | q->required_elevator_features)) | |
1da177e4 LT |
809 | len += sprintf(name+len, "%s ", __e->elevator_name); |
810 | } | |
2a12dcd7 | 811 | spin_unlock(&elv_list_lock); |
1da177e4 | 812 | |
344e9ffc | 813 | if (q->elevator) |
bd166ef1 JA |
814 | len += sprintf(name+len, "none"); |
815 | ||
1da177e4 LT |
816 | len += sprintf(len+name, "\n"); |
817 | return len; | |
818 | } | |
819 | ||
165125e1 JA |
820 | struct request *elv_rb_former_request(struct request_queue *q, |
821 | struct request *rq) | |
2e662b65 JA |
822 | { |
823 | struct rb_node *rbprev = rb_prev(&rq->rb_node); | |
824 | ||
825 | if (rbprev) | |
826 | return rb_entry_rq(rbprev); | |
827 | ||
828 | return NULL; | |
829 | } | |
2e662b65 JA |
830 | EXPORT_SYMBOL(elv_rb_former_request); |
831 | ||
165125e1 JA |
832 | struct request *elv_rb_latter_request(struct request_queue *q, |
833 | struct request *rq) | |
2e662b65 JA |
834 | { |
835 | struct rb_node *rbnext = rb_next(&rq->rb_node); | |
836 | ||
837 | if (rbnext) | |
838 | return rb_entry_rq(rbnext); | |
839 | ||
840 | return NULL; | |
841 | } | |
2e662b65 | 842 | EXPORT_SYMBOL(elv_rb_latter_request); |
f8db3835 JK |
843 | |
844 | static int __init elevator_setup(char *str) | |
845 | { | |
846 | pr_warn("Kernel parameter elevator= does not have any effect anymore.\n" | |
847 | "Please use sysfs to set IO scheduler for individual devices.\n"); | |
848 | return 1; | |
849 | } | |
850 | ||
851 | __setup("elevator=", elevator_setup); |